A Machine Learning based Approach to Detect Sentiment in Twitter Data

This paper presents a machine learning based algorithmic approach to detect sentiment in Tweets posted by
users on microblogging site Twitter. The experimental framework is based on use of a Naïve Bayes classifier.
First of all, the standard Naïve Bayes classifier is implemented in R language and tested on two publicly
available datasets comprising of sentiment labeled tweets. Then the standard Naïve Bayes classifier is
modified to design a Lexicon-pooled hybrid classifier which incorporates knowledge from sentiment lexicon
as well. The designs are evaluated for two feature selection schemes: tf and tf.idf. The accuracy of the
different implementations is calculated and plotted diagrammatically. The proposed approach is a good and
robust approach for detecting sentiment in tweets posted by users.